agentleFS
oeveryoevery/rsp/.agents/skills/author-rsp-skills/SKILL.md

author-rsp-skills

Skill1 starsChanged 8 days agoMITView on GitHub

Maintain RSP's own Skills, source provenance and evaluation evidence. Use for repository Skill maintenance or read-only corpus audits, not general project Skill authoring.

The file

We split this skill into the rules it gives an agent

6 rules. About 0 are common, 5 are distinctive, and 1 is very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈5

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 1

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Read nearest instructions and the selected Change when tracked.

    line 17
  2. Distinctive:

    Do not preload unrelated branches.

    line 34
  3. Distinctive:

    Before editing, identify trigger, inputs, authority, action, output, stops, verification and conditional loading.

    line 40
  4. Distinctive:

    Use host limits as constraints, never as the definition of quality.

    line 42
  5. Distinctive:

    Use rsp-doc's Skill branch for descriptions, instructions, resources and concise expression.

    line 44
  6. Very specific:

    When layout, reachability or repetition matters, use node scripts/scan-skill-context.mjs --package <canonical-path> from the repository …

    line 51

≈ “Common” is an estimate: at least 1 in 5 of the 372,429 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Author RSP Skills

Audit an authored corpus or prepare one bounded RSP Skill candidate. Apply rsp-doc's Skill-authoring method with these RSP-only maintenance requirements. They do not govern ordinary project Skills or grant acceptance, Git, lifecycle or installation authority.

Select mode and target

Require one explicit target package or authored corpus. Use Core's Direct/Change selection; authoring mode alone requires no Change.

  • A report-only Pre-Change Audit is read-only, returns findings with WorkRef: N/A, and stops before candidate creation, repair, mutation or acceptance.
  • Mutation requires explicit artifact authority; tracked work uses its selected Change.

Read nearest instructions and the selected Change when tracked. Preserve unrelated work; unresolved ownership, behavior or authority stops the affected work.

Choose one primary mode:

  • create: a demonstrated capability has no owner.
  • revise: an existing contract or workflow needs a bounded behavior change.
  • audit: the target corpus needs structural, reachability, duplication, or clarity findings.
  • concise: equivalent behavior should use less or clearer context.
  • adapt: an accepted upstream mechanism is selected for local use.
  • evaluate: a candidate needs task evidence, or an explicit comparative claim needs matched runs.

Select reading scope by mode:

  • audit or whole-package concise: inspect complete selected packages, including UI metadata and references.
  • Bounded revise or adapt: read the entrypoint, changed branch, direct callers, relevant tests and adopted research needed by the delta. Expand for shared rules, moved resources or uncertain closure.
  • create: inspect existing neighboring owners. evaluate: read the selected case, candidate and execution inputs.

Do not preload unrelated branches; final metadata/resource validation covers every complete changed package.

Load only the selected reference: authoring for create | revise | audit | adapt, concision for concise, or evaluation for evaluate. For any authorized mutation, including audit repair, additionally load evaluation before completion when observable behavior changes. Report-only audits remain read-only.

Preserve the contract

Before editing, identify trigger, inputs, authority, action, output, stops, verification and conditional loading. Change them only through an authorized delta; record it in the selected Change when tracked. Keep one owner for each state, receipt, field and lifecycle transition.

Use host limits as constraints, never as the definition of quality. Words, lines, bytes, tokens, tool calls, and elapsed time are diagnostics. Do not pass a candidate because it is shorter or fail it because it is longer.

Use rsp-doc's Skill branch for descriptions, instructions, resources and concise expression. If unavailable, use the host's authoring guidance within the same boundary; do not install it implicitly.

Work

  1. Establish current evidence and, when mutation is authorized, the smallest candidate delta. Use the selected comparison baseline; pre-existing work is not part of that delta.
  2. Apply the general authoring method to canonical package sources, not generated projections.
  3. Preserve the declared suite's runtime resource closure; maintainer research and candidate evidence must not become installed dependencies.
  4. When layout, reachability or repetition matters, use node scripts/scan-skill-context.mjs --package <canonical-path> from the repository root. Read context diagnostics for selection and interpretation; scan results are clues, not quality gates.
  5. Reuse repository evaluation, security, packaging and behavior checks. Permitted inspection and writing methods may vary within the same goal, scope, authority, baseline and evidence boundary; observations establish no candidate acceptance. Do not duplicate checks in the Skill or substitute diagnostics for named required operations.
  6. For tracked work, put the authorized candidate delta and approach in Proposal, Spec and Design; record outcomes, evidence, knowledge-update references and limits in Tasks, Verify and Blockers. Keep the durable-decision summary in the response. A report-only Pre-Change Audit returns findings to Core or the user without writing artifacts.

Stop and return

Stop a report-only Pre-Change Audit before any artifact mutation or candidate acceptance. Stop before accepting a candidate, independent review, Git delivery, archive, push, tag, release, publication, or installation unless the user separately authorizes the owning workflow. Also stop when provenance, license, containment, current behavior, or evidence explicitly required by the selected experiment is unresolved. Author self-checks cannot replace required independent review; recommendations for the next owner grant no authority.

Return: WorkRef (N/A for untracked work), Mode, Target, Contract delta, Changed artifacts, Fresh verification, Diagnostics, Blockers, and Next owner. Use natural language; include machine-readable output only when another tool consumes it.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.